Since ChatGPT launched I've been trying to integrate LLMs into my workflows; emails, commits PRSs, or even long chat threads. My instinct was to code a small automation platform, but I found myself limited once again by its own constraints. Went back to zero, twice, and went back to the tools I've been using for years: UNIX pipelines. I started using llm and later mods (now deprecated) and they're great tools, but I wanted something leaner and that would compose better with other tools.Clai does one thing: takes stdin, sends it to a model of your choice and prints the result. No REPL, no state. When it's done it just exits. git diff | clai commit cat article.txt | clai summarize | glow pbpaste | clai tldr curl -s example.com/article.html | clai -e "Extract the three main concepts" It ships with a set of named prompts but you can override, add or extend as needed, they're just files with frontmatter. There's also strategies (chain-of-draft, chain-of-thought, tree-of-thought, and self-re...